2 * Copyright (c) 2016-2017, Mellanox Technologies. All rights reserved.
3 * Copyright (c) 2016-2017, Dave Watson <davejwatson@fb.com>. All rights reserved.
4 * Copyright (c) 2016-2017, Lance Chao <lancerchao@fb.com>. All rights reserved.
5 * Copyright (c) 2016, Fridolin Pokorny <fridolin.pokorny@gmail.com>. All rights reserved.
6 * Copyright (c) 2016, Nikos Mavrogiannopoulos <nmav@gnutls.org>. All rights reserved.
7 * Copyright (c) 2018, Covalent IO, Inc. http://covalent.io
9 * This software is available to you under a choice of one of two
10 * licenses. You may choose to be licensed under the terms of the GNU
11 * General Public License (GPL) Version 2, available from the file
12 * COPYING in the main directory of this source tree, or the
13 * OpenIB.org BSD license below:
15 * Redistribution and use in source and binary forms, with or
16 * without modification, are permitted provided that the following
19 * - Redistributions of source code must retain the above
20 * copyright notice, this list of conditions and the following
23 * - Redistributions in binary form must reproduce the above
24 * copyright notice, this list of conditions and the following
25 * disclaimer in the documentation and/or other materials
26 * provided with the distribution.
28 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
29 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
30 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
31 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
32 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
33 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
34 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
38 #include <linux/bug.h>
39 #include <linux/sched/signal.h>
40 #include <linux/module.h>
41 #include <linux/splice.h>
42 #include <crypto/aead.h>
44 #include <net/strparser.h>
47 noinline void tls_err_abort(struct sock *sk, int err)
49 WARN_ON_ONCE(err >= 0);
50 /* sk->sk_err should contain a positive error code. */
55 static int __skb_nsg(struct sk_buff *skb, int offset, int len,
56 unsigned int recursion_level)
58 int start = skb_headlen(skb);
59 int i, chunk = start - offset;
60 struct sk_buff *frag_iter;
63 if (unlikely(recursion_level >= 24))
76 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
79 WARN_ON(start > offset + len);
81 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
95 if (unlikely(skb_has_frag_list(skb))) {
96 skb_walk_frags(skb, frag_iter) {
99 WARN_ON(start > offset + len);
101 end = start + frag_iter->len;
102 chunk = end - offset;
106 ret = __skb_nsg(frag_iter, offset - start, chunk,
107 recursion_level + 1);
108 if (unlikely(ret < 0))
123 /* Return the number of scatterlist elements required to completely map the
124 * skb, or -EMSGSIZE if the recursion depth is exceeded.
126 static int skb_nsg(struct sk_buff *skb, int offset, int len)
128 return __skb_nsg(skb, offset, len, 0);
131 static int padding_length(struct tls_sw_context_rx *ctx,
132 struct tls_prot_info *prot, struct sk_buff *skb)
134 struct strp_msg *rxm = strp_msg(skb);
137 /* Determine zero-padding length */
138 if (prot->version == TLS_1_3_VERSION) {
139 char content_type = 0;
143 while (content_type == 0) {
144 if (back > rxm->full_len - prot->prepend_size)
146 err = skb_copy_bits(skb,
147 rxm->offset + rxm->full_len - back,
156 ctx->control = content_type;
161 static void tls_decrypt_done(struct crypto_async_request *req, int err)
163 struct aead_request *aead_req = (struct aead_request *)req;
164 struct scatterlist *sgout = aead_req->dst;
165 struct scatterlist *sgin = aead_req->src;
166 struct tls_sw_context_rx *ctx;
167 struct tls_context *tls_ctx;
168 struct tls_prot_info *prot;
169 struct scatterlist *sg;
174 skb = (struct sk_buff *)req->data;
175 tls_ctx = tls_get_ctx(skb->sk);
176 ctx = tls_sw_ctx_rx(tls_ctx);
177 prot = &tls_ctx->prot_info;
179 /* Propagate if there was an err */
182 TLS_INC_STATS(sock_net(skb->sk),
183 LINUX_MIB_TLSDECRYPTERROR);
184 ctx->async_wait.err = err;
185 tls_err_abort(skb->sk, err);
187 struct strp_msg *rxm = strp_msg(skb);
190 pad = padding_length(ctx, prot, skb);
192 ctx->async_wait.err = pad;
193 tls_err_abort(skb->sk, pad);
195 rxm->full_len -= pad;
196 rxm->offset += prot->prepend_size;
197 rxm->full_len -= prot->overhead_size;
201 /* After using skb->sk to propagate sk through crypto async callback
202 * we need to NULL it again.
207 /* Free the destination pages if skb was not decrypted inplace */
209 /* Skip the first S/G entry as it points to AAD */
210 for_each_sg(sg_next(sgout), sg, UINT_MAX, pages) {
213 put_page(sg_page(sg));
219 spin_lock_bh(&ctx->decrypt_compl_lock);
220 pending = atomic_dec_return(&ctx->decrypt_pending);
222 if (!pending && ctx->async_notify)
223 complete(&ctx->async_wait.completion);
224 spin_unlock_bh(&ctx->decrypt_compl_lock);
227 static int tls_do_decryption(struct sock *sk,
229 struct scatterlist *sgin,
230 struct scatterlist *sgout,
233 struct aead_request *aead_req,
236 struct tls_context *tls_ctx = tls_get_ctx(sk);
237 struct tls_prot_info *prot = &tls_ctx->prot_info;
238 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
241 aead_request_set_tfm(aead_req, ctx->aead_recv);
242 aead_request_set_ad(aead_req, prot->aad_size);
243 aead_request_set_crypt(aead_req, sgin, sgout,
244 data_len + prot->tag_size,
248 /* Using skb->sk to push sk through to crypto async callback
249 * handler. This allows propagating errors up to the socket
250 * if needed. It _must_ be cleared in the async handler
251 * before consume_skb is called. We _know_ skb->sk is NULL
252 * because it is a clone from strparser.
255 aead_request_set_callback(aead_req,
256 CRYPTO_TFM_REQ_MAY_BACKLOG,
257 tls_decrypt_done, skb);
258 atomic_inc(&ctx->decrypt_pending);
260 aead_request_set_callback(aead_req,
261 CRYPTO_TFM_REQ_MAY_BACKLOG,
262 crypto_req_done, &ctx->async_wait);
265 ret = crypto_aead_decrypt(aead_req);
266 if (ret == -EINPROGRESS) {
270 ret = crypto_wait_req(ret, &ctx->async_wait);
274 atomic_dec(&ctx->decrypt_pending);
279 static void tls_trim_both_msgs(struct sock *sk, int target_size)
281 struct tls_context *tls_ctx = tls_get_ctx(sk);
282 struct tls_prot_info *prot = &tls_ctx->prot_info;
283 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
284 struct tls_rec *rec = ctx->open_rec;
286 sk_msg_trim(sk, &rec->msg_plaintext, target_size);
288 target_size += prot->overhead_size;
289 sk_msg_trim(sk, &rec->msg_encrypted, target_size);
292 static int tls_alloc_encrypted_msg(struct sock *sk, int len)
294 struct tls_context *tls_ctx = tls_get_ctx(sk);
295 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
296 struct tls_rec *rec = ctx->open_rec;
297 struct sk_msg *msg_en = &rec->msg_encrypted;
299 return sk_msg_alloc(sk, msg_en, len, 0);
302 static int tls_clone_plaintext_msg(struct sock *sk, int required)
304 struct tls_context *tls_ctx = tls_get_ctx(sk);
305 struct tls_prot_info *prot = &tls_ctx->prot_info;
306 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
307 struct tls_rec *rec = ctx->open_rec;
308 struct sk_msg *msg_pl = &rec->msg_plaintext;
309 struct sk_msg *msg_en = &rec->msg_encrypted;
312 /* We add page references worth len bytes from encrypted sg
313 * at the end of plaintext sg. It is guaranteed that msg_en
314 * has enough required room (ensured by caller).
316 len = required - msg_pl->sg.size;
318 /* Skip initial bytes in msg_en's data to be able to use
319 * same offset of both plain and encrypted data.
321 skip = prot->prepend_size + msg_pl->sg.size;
323 return sk_msg_clone(sk, msg_pl, msg_en, skip, len);
326 static struct tls_rec *tls_get_rec(struct sock *sk)
328 struct tls_context *tls_ctx = tls_get_ctx(sk);
329 struct tls_prot_info *prot = &tls_ctx->prot_info;
330 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
331 struct sk_msg *msg_pl, *msg_en;
335 mem_size = sizeof(struct tls_rec) + crypto_aead_reqsize(ctx->aead_send);
337 rec = kzalloc(mem_size, sk->sk_allocation);
341 msg_pl = &rec->msg_plaintext;
342 msg_en = &rec->msg_encrypted;
347 sg_init_table(rec->sg_aead_in, 2);
348 sg_set_buf(&rec->sg_aead_in[0], rec->aad_space, prot->aad_size);
349 sg_unmark_end(&rec->sg_aead_in[1]);
351 sg_init_table(rec->sg_aead_out, 2);
352 sg_set_buf(&rec->sg_aead_out[0], rec->aad_space, prot->aad_size);
353 sg_unmark_end(&rec->sg_aead_out[1]);
358 static void tls_free_rec(struct sock *sk, struct tls_rec *rec)
360 sk_msg_free(sk, &rec->msg_encrypted);
361 sk_msg_free(sk, &rec->msg_plaintext);
365 static void tls_free_open_rec(struct sock *sk)
367 struct tls_context *tls_ctx = tls_get_ctx(sk);
368 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
369 struct tls_rec *rec = ctx->open_rec;
372 tls_free_rec(sk, rec);
373 ctx->open_rec = NULL;
377 int tls_tx_records(struct sock *sk, int flags)
379 struct tls_context *tls_ctx = tls_get_ctx(sk);
380 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
381 struct tls_rec *rec, *tmp;
382 struct sk_msg *msg_en;
383 int tx_flags, rc = 0;
385 if (tls_is_partially_sent_record(tls_ctx)) {
386 rec = list_first_entry(&ctx->tx_list,
387 struct tls_rec, list);
390 tx_flags = rec->tx_flags;
394 rc = tls_push_partial_record(sk, tls_ctx, tx_flags);
398 /* Full record has been transmitted.
399 * Remove the head of tx_list
401 list_del(&rec->list);
402 sk_msg_free(sk, &rec->msg_plaintext);
406 /* Tx all ready records */
407 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
408 if (READ_ONCE(rec->tx_ready)) {
410 tx_flags = rec->tx_flags;
414 msg_en = &rec->msg_encrypted;
415 rc = tls_push_sg(sk, tls_ctx,
416 &msg_en->sg.data[msg_en->sg.curr],
421 list_del(&rec->list);
422 sk_msg_free(sk, &rec->msg_plaintext);
430 if (rc < 0 && rc != -EAGAIN)
431 tls_err_abort(sk, -EBADMSG);
436 static void tls_encrypt_done(struct crypto_async_request *req, int err)
438 struct aead_request *aead_req = (struct aead_request *)req;
439 struct sock *sk = req->data;
440 struct tls_context *tls_ctx = tls_get_ctx(sk);
441 struct tls_prot_info *prot = &tls_ctx->prot_info;
442 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
443 struct scatterlist *sge;
444 struct sk_msg *msg_en;
449 rec = container_of(aead_req, struct tls_rec, aead_req);
450 msg_en = &rec->msg_encrypted;
452 sge = sk_msg_elem(msg_en, msg_en->sg.curr);
453 sge->offset -= prot->prepend_size;
454 sge->length += prot->prepend_size;
456 /* Check if error is previously set on socket */
457 if (err || sk->sk_err) {
460 /* If err is already set on socket, return the same code */
462 ctx->async_wait.err = -sk->sk_err;
464 ctx->async_wait.err = err;
465 tls_err_abort(sk, err);
470 struct tls_rec *first_rec;
472 /* Mark the record as ready for transmission */
473 smp_store_mb(rec->tx_ready, true);
475 /* If received record is at head of tx_list, schedule tx */
476 first_rec = list_first_entry(&ctx->tx_list,
477 struct tls_rec, list);
478 if (rec == first_rec)
482 spin_lock_bh(&ctx->encrypt_compl_lock);
483 pending = atomic_dec_return(&ctx->encrypt_pending);
485 if (!pending && ctx->async_notify)
486 complete(&ctx->async_wait.completion);
487 spin_unlock_bh(&ctx->encrypt_compl_lock);
492 /* Schedule the transmission */
493 if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
494 schedule_delayed_work(&ctx->tx_work.work, 1);
497 static int tls_do_encryption(struct sock *sk,
498 struct tls_context *tls_ctx,
499 struct tls_sw_context_tx *ctx,
500 struct aead_request *aead_req,
501 size_t data_len, u32 start)
503 struct tls_prot_info *prot = &tls_ctx->prot_info;
504 struct tls_rec *rec = ctx->open_rec;
505 struct sk_msg *msg_en = &rec->msg_encrypted;
506 struct scatterlist *sge = sk_msg_elem(msg_en, start);
507 int rc, iv_offset = 0;
509 /* For CCM based ciphers, first byte of IV is a constant */
510 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
511 rec->iv_data[0] = TLS_AES_CCM_IV_B0_BYTE;
515 memcpy(&rec->iv_data[iv_offset], tls_ctx->tx.iv,
516 prot->iv_size + prot->salt_size);
518 xor_iv_with_seq(prot, rec->iv_data + iv_offset, tls_ctx->tx.rec_seq);
520 sge->offset += prot->prepend_size;
521 sge->length -= prot->prepend_size;
523 msg_en->sg.curr = start;
525 aead_request_set_tfm(aead_req, ctx->aead_send);
526 aead_request_set_ad(aead_req, prot->aad_size);
527 aead_request_set_crypt(aead_req, rec->sg_aead_in,
529 data_len, rec->iv_data);
531 aead_request_set_callback(aead_req, CRYPTO_TFM_REQ_MAY_BACKLOG,
532 tls_encrypt_done, sk);
534 /* Add the record in tx_list */
535 list_add_tail((struct list_head *)&rec->list, &ctx->tx_list);
536 atomic_inc(&ctx->encrypt_pending);
538 rc = crypto_aead_encrypt(aead_req);
539 if (!rc || rc != -EINPROGRESS) {
540 atomic_dec(&ctx->encrypt_pending);
541 sge->offset -= prot->prepend_size;
542 sge->length += prot->prepend_size;
546 WRITE_ONCE(rec->tx_ready, true);
547 } else if (rc != -EINPROGRESS) {
548 list_del(&rec->list);
552 /* Unhook the record from context if encryption is not failure */
553 ctx->open_rec = NULL;
554 tls_advance_record_sn(sk, prot, &tls_ctx->tx);
558 static int tls_split_open_record(struct sock *sk, struct tls_rec *from,
559 struct tls_rec **to, struct sk_msg *msg_opl,
560 struct sk_msg *msg_oen, u32 split_point,
561 u32 tx_overhead_size, u32 *orig_end)
563 u32 i, j, bytes = 0, apply = msg_opl->apply_bytes;
564 struct scatterlist *sge, *osge, *nsge;
565 u32 orig_size = msg_opl->sg.size;
566 struct scatterlist tmp = { };
567 struct sk_msg *msg_npl;
571 new = tls_get_rec(sk);
574 ret = sk_msg_alloc(sk, &new->msg_encrypted, msg_opl->sg.size +
575 tx_overhead_size, 0);
577 tls_free_rec(sk, new);
581 *orig_end = msg_opl->sg.end;
582 i = msg_opl->sg.start;
583 sge = sk_msg_elem(msg_opl, i);
584 while (apply && sge->length) {
585 if (sge->length > apply) {
586 u32 len = sge->length - apply;
588 get_page(sg_page(sge));
589 sg_set_page(&tmp, sg_page(sge), len,
590 sge->offset + apply);
595 apply -= sge->length;
596 bytes += sge->length;
599 sk_msg_iter_var_next(i);
600 if (i == msg_opl->sg.end)
602 sge = sk_msg_elem(msg_opl, i);
606 msg_opl->sg.curr = i;
607 msg_opl->sg.copybreak = 0;
608 msg_opl->apply_bytes = 0;
609 msg_opl->sg.size = bytes;
611 msg_npl = &new->msg_plaintext;
612 msg_npl->apply_bytes = apply;
613 msg_npl->sg.size = orig_size - bytes;
615 j = msg_npl->sg.start;
616 nsge = sk_msg_elem(msg_npl, j);
618 memcpy(nsge, &tmp, sizeof(*nsge));
619 sk_msg_iter_var_next(j);
620 nsge = sk_msg_elem(msg_npl, j);
623 osge = sk_msg_elem(msg_opl, i);
624 while (osge->length) {
625 memcpy(nsge, osge, sizeof(*nsge));
627 sk_msg_iter_var_next(i);
628 sk_msg_iter_var_next(j);
631 osge = sk_msg_elem(msg_opl, i);
632 nsge = sk_msg_elem(msg_npl, j);
636 msg_npl->sg.curr = j;
637 msg_npl->sg.copybreak = 0;
643 static void tls_merge_open_record(struct sock *sk, struct tls_rec *to,
644 struct tls_rec *from, u32 orig_end)
646 struct sk_msg *msg_npl = &from->msg_plaintext;
647 struct sk_msg *msg_opl = &to->msg_plaintext;
648 struct scatterlist *osge, *nsge;
652 sk_msg_iter_var_prev(i);
653 j = msg_npl->sg.start;
655 osge = sk_msg_elem(msg_opl, i);
656 nsge = sk_msg_elem(msg_npl, j);
658 if (sg_page(osge) == sg_page(nsge) &&
659 osge->offset + osge->length == nsge->offset) {
660 osge->length += nsge->length;
661 put_page(sg_page(nsge));
664 msg_opl->sg.end = orig_end;
665 msg_opl->sg.curr = orig_end;
666 msg_opl->sg.copybreak = 0;
667 msg_opl->apply_bytes = msg_opl->sg.size + msg_npl->sg.size;
668 msg_opl->sg.size += msg_npl->sg.size;
670 sk_msg_free(sk, &to->msg_encrypted);
671 sk_msg_xfer_full(&to->msg_encrypted, &from->msg_encrypted);
676 static int tls_push_record(struct sock *sk, int flags,
677 unsigned char record_type)
679 struct tls_context *tls_ctx = tls_get_ctx(sk);
680 struct tls_prot_info *prot = &tls_ctx->prot_info;
681 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
682 struct tls_rec *rec = ctx->open_rec, *tmp = NULL;
683 u32 i, split_point, orig_end;
684 struct sk_msg *msg_pl, *msg_en;
685 struct aead_request *req;
692 msg_pl = &rec->msg_plaintext;
693 msg_en = &rec->msg_encrypted;
695 split_point = msg_pl->apply_bytes;
696 split = split_point && split_point < msg_pl->sg.size;
697 if (unlikely((!split &&
699 prot->overhead_size > msg_en->sg.size) ||
702 prot->overhead_size > msg_en->sg.size))) {
704 split_point = msg_en->sg.size;
707 rc = tls_split_open_record(sk, rec, &tmp, msg_pl, msg_en,
708 split_point, prot->overhead_size,
712 /* This can happen if above tls_split_open_record allocates
713 * a single large encryption buffer instead of two smaller
714 * ones. In this case adjust pointers and continue without
717 if (!msg_pl->sg.size) {
718 tls_merge_open_record(sk, rec, tmp, orig_end);
719 msg_pl = &rec->msg_plaintext;
720 msg_en = &rec->msg_encrypted;
723 sk_msg_trim(sk, msg_en, msg_pl->sg.size +
724 prot->overhead_size);
727 rec->tx_flags = flags;
728 req = &rec->aead_req;
731 sk_msg_iter_var_prev(i);
733 rec->content_type = record_type;
734 if (prot->version == TLS_1_3_VERSION) {
735 /* Add content type to end of message. No padding added */
736 sg_set_buf(&rec->sg_content_type, &rec->content_type, 1);
737 sg_mark_end(&rec->sg_content_type);
738 sg_chain(msg_pl->sg.data, msg_pl->sg.end + 1,
739 &rec->sg_content_type);
741 sg_mark_end(sk_msg_elem(msg_pl, i));
744 if (msg_pl->sg.end < msg_pl->sg.start) {
745 sg_chain(&msg_pl->sg.data[msg_pl->sg.start],
746 MAX_SKB_FRAGS - msg_pl->sg.start + 1,
750 i = msg_pl->sg.start;
751 sg_chain(rec->sg_aead_in, 2, &msg_pl->sg.data[i]);
754 sk_msg_iter_var_prev(i);
755 sg_mark_end(sk_msg_elem(msg_en, i));
757 i = msg_en->sg.start;
758 sg_chain(rec->sg_aead_out, 2, &msg_en->sg.data[i]);
760 tls_make_aad(rec->aad_space, msg_pl->sg.size + prot->tail_size,
761 tls_ctx->tx.rec_seq, record_type, prot);
763 tls_fill_prepend(tls_ctx,
764 page_address(sg_page(&msg_en->sg.data[i])) +
765 msg_en->sg.data[i].offset,
766 msg_pl->sg.size + prot->tail_size,
769 tls_ctx->pending_open_record_frags = false;
771 rc = tls_do_encryption(sk, tls_ctx, ctx, req,
772 msg_pl->sg.size + prot->tail_size, i);
774 if (rc != -EINPROGRESS) {
775 tls_err_abort(sk, -EBADMSG);
777 tls_ctx->pending_open_record_frags = true;
778 tls_merge_open_record(sk, rec, tmp, orig_end);
781 ctx->async_capable = 1;
784 msg_pl = &tmp->msg_plaintext;
785 msg_en = &tmp->msg_encrypted;
786 sk_msg_trim(sk, msg_en, msg_pl->sg.size + prot->overhead_size);
787 tls_ctx->pending_open_record_frags = true;
791 return tls_tx_records(sk, flags);
794 static int bpf_exec_tx_verdict(struct sk_msg *msg, struct sock *sk,
795 bool full_record, u8 record_type,
796 ssize_t *copied, int flags)
798 struct tls_context *tls_ctx = tls_get_ctx(sk);
799 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
800 struct sk_msg msg_redir = { };
801 struct sk_psock *psock;
802 struct sock *sk_redir;
804 bool enospc, policy, redir_ingress;
808 policy = !(flags & MSG_SENDPAGE_NOPOLICY);
809 psock = sk_psock_get(sk);
810 if (!psock || !policy) {
811 err = tls_push_record(sk, flags, record_type);
812 if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
813 *copied -= sk_msg_free(sk, msg);
814 tls_free_open_rec(sk);
818 sk_psock_put(sk, psock);
822 enospc = sk_msg_full(msg);
823 if (psock->eval == __SK_NONE) {
824 delta = msg->sg.size;
825 psock->eval = sk_psock_msg_verdict(sk, psock, msg);
826 delta -= msg->sg.size;
828 if (msg->cork_bytes && msg->cork_bytes > msg->sg.size &&
829 !enospc && !full_record) {
835 if (msg->apply_bytes && msg->apply_bytes < send)
836 send = msg->apply_bytes;
838 switch (psock->eval) {
840 err = tls_push_record(sk, flags, record_type);
841 if (err && err != -EINPROGRESS && sk->sk_err == EBADMSG) {
842 *copied -= sk_msg_free(sk, msg);
843 tls_free_open_rec(sk);
849 redir_ingress = psock->redir_ingress;
850 sk_redir = psock->sk_redir;
851 memcpy(&msg_redir, msg, sizeof(*msg));
852 if (msg->apply_bytes < send)
853 msg->apply_bytes = 0;
855 msg->apply_bytes -= send;
856 sk_msg_return_zero(sk, msg, send);
857 msg->sg.size -= send;
859 err = tcp_bpf_sendmsg_redir(sk_redir, redir_ingress,
860 &msg_redir, send, flags);
863 *copied -= sk_msg_free_nocharge(sk, &msg_redir);
866 if (msg->sg.size == 0)
867 tls_free_open_rec(sk);
871 sk_msg_free_partial(sk, msg, send);
872 if (msg->apply_bytes < send)
873 msg->apply_bytes = 0;
875 msg->apply_bytes -= send;
876 if (msg->sg.size == 0)
877 tls_free_open_rec(sk);
878 *copied -= (send + delta);
883 bool reset_eval = !ctx->open_rec;
887 msg = &rec->msg_plaintext;
888 if (!msg->apply_bytes)
892 psock->eval = __SK_NONE;
893 if (psock->sk_redir) {
894 sock_put(psock->sk_redir);
895 psock->sk_redir = NULL;
902 sk_psock_put(sk, psock);
906 static int tls_sw_push_pending_record(struct sock *sk, int flags)
908 struct tls_context *tls_ctx = tls_get_ctx(sk);
909 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
910 struct tls_rec *rec = ctx->open_rec;
911 struct sk_msg *msg_pl;
917 msg_pl = &rec->msg_plaintext;
918 copied = msg_pl->sg.size;
922 return bpf_exec_tx_verdict(msg_pl, sk, true, TLS_RECORD_TYPE_DATA,
926 int tls_sw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
928 long timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
929 struct tls_context *tls_ctx = tls_get_ctx(sk);
930 struct tls_prot_info *prot = &tls_ctx->prot_info;
931 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
932 bool async_capable = ctx->async_capable;
933 unsigned char record_type = TLS_RECORD_TYPE_DATA;
934 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
935 bool eor = !(msg->msg_flags & MSG_MORE);
938 struct sk_msg *msg_pl, *msg_en;
949 if (msg->msg_flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
953 ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
958 if (unlikely(msg->msg_controllen)) {
959 ret = tls_proccess_cmsg(sk, msg, &record_type);
961 if (ret == -EINPROGRESS)
963 else if (ret != -EAGAIN)
968 while (msg_data_left(msg)) {
977 rec = ctx->open_rec = tls_get_rec(sk);
983 msg_pl = &rec->msg_plaintext;
984 msg_en = &rec->msg_encrypted;
986 orig_size = msg_pl->sg.size;
988 try_to_copy = msg_data_left(msg);
989 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
990 if (try_to_copy >= record_room) {
991 try_to_copy = record_room;
995 required_size = msg_pl->sg.size + try_to_copy +
998 if (!sk_stream_memory_free(sk))
999 goto wait_for_sndbuf;
1002 ret = tls_alloc_encrypted_msg(sk, required_size);
1005 goto wait_for_memory;
1007 /* Adjust try_to_copy according to the amount that was
1008 * actually allocated. The difference is due
1009 * to max sg elements limit
1011 try_to_copy -= required_size - msg_en->sg.size;
1015 if (!is_kvec && (full_record || eor) && !async_capable) {
1016 u32 first = msg_pl->sg.end;
1018 ret = sk_msg_zerocopy_from_iter(sk, &msg->msg_iter,
1019 msg_pl, try_to_copy);
1021 goto fallback_to_reg_send;
1024 copied += try_to_copy;
1026 sk_msg_sg_copy_set(msg_pl, first);
1027 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1028 record_type, &copied,
1031 if (ret == -EINPROGRESS)
1033 else if (ret == -ENOMEM)
1034 goto wait_for_memory;
1035 else if (ctx->open_rec && ret == -ENOSPC)
1037 else if (ret != -EAGAIN)
1042 copied -= try_to_copy;
1043 sk_msg_sg_copy_clear(msg_pl, first);
1044 iov_iter_revert(&msg->msg_iter,
1045 msg_pl->sg.size - orig_size);
1046 fallback_to_reg_send:
1047 sk_msg_trim(sk, msg_pl, orig_size);
1050 required_size = msg_pl->sg.size + try_to_copy;
1052 ret = tls_clone_plaintext_msg(sk, required_size);
1057 /* Adjust try_to_copy according to the amount that was
1058 * actually allocated. The difference is due
1059 * to max sg elements limit
1061 try_to_copy -= required_size - msg_pl->sg.size;
1063 sk_msg_trim(sk, msg_en,
1064 msg_pl->sg.size + prot->overhead_size);
1068 ret = sk_msg_memcopy_from_iter(sk, &msg->msg_iter,
1069 msg_pl, try_to_copy);
1074 /* Open records defined only if successfully copied, otherwise
1075 * we would trim the sg but not reset the open record frags.
1077 tls_ctx->pending_open_record_frags = true;
1078 copied += try_to_copy;
1079 if (full_record || eor) {
1080 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1081 record_type, &copied,
1084 if (ret == -EINPROGRESS)
1086 else if (ret == -ENOMEM)
1087 goto wait_for_memory;
1088 else if (ret != -EAGAIN) {
1099 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1101 ret = sk_stream_wait_memory(sk, &timeo);
1105 tls_trim_both_msgs(sk, orig_size);
1109 if (ctx->open_rec && msg_en->sg.size < required_size)
1110 goto alloc_encrypted;
1115 } else if (num_zc) {
1116 /* Wait for pending encryptions to get completed */
1117 spin_lock_bh(&ctx->encrypt_compl_lock);
1118 ctx->async_notify = true;
1120 pending = atomic_read(&ctx->encrypt_pending);
1121 spin_unlock_bh(&ctx->encrypt_compl_lock);
1123 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1125 reinit_completion(&ctx->async_wait.completion);
1127 /* There can be no concurrent accesses, since we have no
1128 * pending encrypt operations
1130 WRITE_ONCE(ctx->async_notify, false);
1132 if (ctx->async_wait.err) {
1133 ret = ctx->async_wait.err;
1138 /* Transmit if any encryptions have completed */
1139 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1140 cancel_delayed_work(&ctx->tx_work.work);
1141 tls_tx_records(sk, msg->msg_flags);
1145 ret = sk_stream_error(sk, msg->msg_flags, ret);
1148 mutex_unlock(&tls_ctx->tx_lock);
1149 return copied > 0 ? copied : ret;
1152 static int tls_sw_do_sendpage(struct sock *sk, struct page *page,
1153 int offset, size_t size, int flags)
1155 long timeo = sock_sndtimeo(sk, flags & MSG_DONTWAIT);
1156 struct tls_context *tls_ctx = tls_get_ctx(sk);
1157 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
1158 struct tls_prot_info *prot = &tls_ctx->prot_info;
1159 unsigned char record_type = TLS_RECORD_TYPE_DATA;
1160 struct sk_msg *msg_pl;
1161 struct tls_rec *rec;
1169 eor = !(flags & MSG_SENDPAGE_NOTLAST);
1170 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1172 /* Call the sk_stream functions to manage the sndbuf mem. */
1174 size_t copy, required_size;
1182 rec = ctx->open_rec;
1184 rec = ctx->open_rec = tls_get_rec(sk);
1190 msg_pl = &rec->msg_plaintext;
1192 full_record = false;
1193 record_room = TLS_MAX_PAYLOAD_SIZE - msg_pl->sg.size;
1195 if (copy >= record_room) {
1200 required_size = msg_pl->sg.size + copy + prot->overhead_size;
1202 if (!sk_stream_memory_free(sk))
1203 goto wait_for_sndbuf;
1205 ret = tls_alloc_encrypted_msg(sk, required_size);
1208 goto wait_for_memory;
1210 /* Adjust copy according to the amount that was
1211 * actually allocated. The difference is due
1212 * to max sg elements limit
1214 copy -= required_size - msg_pl->sg.size;
1218 sk_msg_page_add(msg_pl, page, copy, offset);
1219 sk_mem_charge(sk, copy);
1225 tls_ctx->pending_open_record_frags = true;
1226 if (full_record || eor || sk_msg_full(msg_pl)) {
1227 ret = bpf_exec_tx_verdict(msg_pl, sk, full_record,
1228 record_type, &copied, flags);
1230 if (ret == -EINPROGRESS)
1232 else if (ret == -ENOMEM)
1233 goto wait_for_memory;
1234 else if (ret != -EAGAIN) {
1243 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1245 ret = sk_stream_wait_memory(sk, &timeo);
1248 tls_trim_both_msgs(sk, msg_pl->sg.size);
1257 /* Transmit if any encryptions have completed */
1258 if (test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
1259 cancel_delayed_work(&ctx->tx_work.work);
1260 tls_tx_records(sk, flags);
1264 ret = sk_stream_error(sk, flags, ret);
1265 return copied > 0 ? copied : ret;
1268 int tls_sw_sendpage_locked(struct sock *sk, struct page *page,
1269 int offset, size_t size, int flags)
1271 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1272 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY |
1273 MSG_NO_SHARED_FRAGS))
1276 return tls_sw_do_sendpage(sk, page, offset, size, flags);
1279 int tls_sw_sendpage(struct sock *sk, struct page *page,
1280 int offset, size_t size, int flags)
1282 struct tls_context *tls_ctx = tls_get_ctx(sk);
1285 if (flags & ~(MSG_MORE | MSG_DONTWAIT | MSG_NOSIGNAL |
1286 MSG_SENDPAGE_NOTLAST | MSG_SENDPAGE_NOPOLICY))
1289 ret = mutex_lock_interruptible(&tls_ctx->tx_lock);
1293 ret = tls_sw_do_sendpage(sk, page, offset, size, flags);
1295 mutex_unlock(&tls_ctx->tx_lock);
1299 static struct sk_buff *tls_wait_data(struct sock *sk, struct sk_psock *psock,
1300 bool nonblock, long timeo, int *err)
1302 struct tls_context *tls_ctx = tls_get_ctx(sk);
1303 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1304 struct sk_buff *skb;
1305 DEFINE_WAIT_FUNC(wait, woken_wake_function);
1307 while (!(skb = ctx->recv_pkt) && sk_psock_queue_empty(psock)) {
1309 *err = sock_error(sk);
1313 if (!skb_queue_empty(&sk->sk_receive_queue)) {
1314 __strp_unpause(&ctx->strp);
1316 return ctx->recv_pkt;
1319 if (sk->sk_shutdown & RCV_SHUTDOWN)
1322 if (sock_flag(sk, SOCK_DONE))
1325 if (nonblock || !timeo) {
1330 add_wait_queue(sk_sleep(sk), &wait);
1331 sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1332 sk_wait_event(sk, &timeo,
1333 ctx->recv_pkt != skb ||
1334 !sk_psock_queue_empty(psock),
1336 sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
1337 remove_wait_queue(sk_sleep(sk), &wait);
1339 /* Handle signals */
1340 if (signal_pending(current)) {
1341 *err = sock_intr_errno(timeo);
1349 static int tls_setup_from_iter(struct sock *sk, struct iov_iter *from,
1350 int length, int *pages_used,
1351 unsigned int *size_used,
1352 struct scatterlist *to,
1355 int rc = 0, i = 0, num_elem = *pages_used, maxpages;
1356 struct page *pages[MAX_SKB_FRAGS];
1357 unsigned int size = *size_used;
1358 ssize_t copied, use;
1361 while (length > 0) {
1363 maxpages = to_max_pages - num_elem;
1364 if (maxpages == 0) {
1368 copied = iov_iter_get_pages(from, pages,
1376 iov_iter_advance(from, copied);
1381 use = min_t(int, copied, PAGE_SIZE - offset);
1383 sg_set_page(&to[num_elem],
1384 pages[i], use, offset);
1385 sg_unmark_end(&to[num_elem]);
1386 /* We do not uncharge memory from this API */
1395 /* Mark the end in the last sg entry if newly added */
1396 if (num_elem > *pages_used)
1397 sg_mark_end(&to[num_elem - 1]);
1400 iov_iter_revert(from, size - *size_used);
1402 *pages_used = num_elem;
1407 /* This function decrypts the input skb into either out_iov or in out_sg
1408 * or in skb buffers itself. The input parameter 'zc' indicates if
1409 * zero-copy mode needs to be tried or not. With zero-copy mode, either
1410 * out_iov or out_sg must be non-NULL. In case both out_iov and out_sg are
1411 * NULL, then the decryption happens inside skb buffers itself, i.e.
1412 * zero-copy gets disabled and 'zc' is updated.
1415 static int decrypt_internal(struct sock *sk, struct sk_buff *skb,
1416 struct iov_iter *out_iov,
1417 struct scatterlist *out_sg,
1418 int *chunk, bool *zc, bool async)
1420 struct tls_context *tls_ctx = tls_get_ctx(sk);
1421 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1422 struct tls_prot_info *prot = &tls_ctx->prot_info;
1423 struct strp_msg *rxm = strp_msg(skb);
1424 int n_sgin, n_sgout, nsg, mem_size, aead_size, err, pages = 0;
1425 struct aead_request *aead_req;
1426 struct sk_buff *unused;
1427 u8 *aad, *iv, *mem = NULL;
1428 struct scatterlist *sgin = NULL;
1429 struct scatterlist *sgout = NULL;
1430 const int data_len = rxm->full_len - prot->overhead_size +
1434 if (*zc && (out_iov || out_sg)) {
1436 n_sgout = iov_iter_npages(out_iov, INT_MAX) + 1;
1438 n_sgout = sg_nents(out_sg);
1439 n_sgin = skb_nsg(skb, rxm->offset + prot->prepend_size,
1440 rxm->full_len - prot->prepend_size);
1444 n_sgin = skb_cow_data(skb, 0, &unused);
1450 /* Increment to accommodate AAD */
1451 n_sgin = n_sgin + 1;
1453 nsg = n_sgin + n_sgout;
1455 aead_size = sizeof(*aead_req) + crypto_aead_reqsize(ctx->aead_recv);
1456 mem_size = aead_size + (nsg * sizeof(struct scatterlist));
1457 mem_size = mem_size + prot->aad_size;
1458 mem_size = mem_size + crypto_aead_ivsize(ctx->aead_recv);
1460 /* Allocate a single block of memory which contains
1461 * aead_req || sgin[] || sgout[] || aad || iv.
1462 * This order achieves correct alignment for aead_req, sgin, sgout.
1464 mem = kmalloc(mem_size, sk->sk_allocation);
1468 /* Segment the allocated memory */
1469 aead_req = (struct aead_request *)mem;
1470 sgin = (struct scatterlist *)(mem + aead_size);
1471 sgout = sgin + n_sgin;
1472 aad = (u8 *)(sgout + n_sgout);
1473 iv = aad + prot->aad_size;
1475 /* For CCM based ciphers, first byte of nonce+iv is always '2' */
1476 if (prot->cipher_type == TLS_CIPHER_AES_CCM_128) {
1482 err = skb_copy_bits(skb, rxm->offset + TLS_HEADER_SIZE,
1483 iv + iv_offset + prot->salt_size,
1489 if (prot->version == TLS_1_3_VERSION ||
1490 prot->cipher_type == TLS_CIPHER_CHACHA20_POLY1305)
1491 memcpy(iv + iv_offset, tls_ctx->rx.iv,
1492 prot->iv_size + prot->salt_size);
1494 memcpy(iv + iv_offset, tls_ctx->rx.iv, prot->salt_size);
1496 xor_iv_with_seq(prot, iv + iv_offset, tls_ctx->rx.rec_seq);
1499 tls_make_aad(aad, rxm->full_len - prot->overhead_size +
1501 tls_ctx->rx.rec_seq, ctx->control, prot);
1504 sg_init_table(sgin, n_sgin);
1505 sg_set_buf(&sgin[0], aad, prot->aad_size);
1506 err = skb_to_sgvec(skb, &sgin[1],
1507 rxm->offset + prot->prepend_size,
1508 rxm->full_len - prot->prepend_size);
1516 sg_init_table(sgout, n_sgout);
1517 sg_set_buf(&sgout[0], aad, prot->aad_size);
1520 err = tls_setup_from_iter(sk, out_iov, data_len,
1521 &pages, chunk, &sgout[1],
1524 goto fallback_to_reg_recv;
1525 } else if (out_sg) {
1526 memcpy(sgout, out_sg, n_sgout * sizeof(*sgout));
1528 goto fallback_to_reg_recv;
1531 fallback_to_reg_recv:
1538 /* Prepare and submit AEAD request */
1539 err = tls_do_decryption(sk, skb, sgin, sgout, iv,
1540 data_len, aead_req, async);
1541 if (err == -EINPROGRESS)
1544 /* Release the pages in case iov was mapped to pages */
1545 for (; pages > 0; pages--)
1546 put_page(sg_page(&sgout[pages]));
1552 static int decrypt_skb_update(struct sock *sk, struct sk_buff *skb,
1553 struct iov_iter *dest, int *chunk, bool *zc,
1556 struct tls_context *tls_ctx = tls_get_ctx(sk);
1557 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1558 struct tls_prot_info *prot = &tls_ctx->prot_info;
1559 struct strp_msg *rxm = strp_msg(skb);
1562 if (!ctx->decrypted) {
1563 if (tls_ctx->rx_conf == TLS_HW) {
1564 err = tls_device_decrypted(sk, tls_ctx, skb, rxm);
1569 /* Still not decrypted after tls_device */
1570 if (!ctx->decrypted) {
1571 err = decrypt_internal(sk, skb, dest, NULL, chunk, zc,
1574 if (err == -EINPROGRESS)
1575 tls_advance_record_sn(sk, prot,
1577 else if (err == -EBADMSG)
1578 TLS_INC_STATS(sock_net(sk),
1579 LINUX_MIB_TLSDECRYPTERROR);
1586 pad = padding_length(ctx, prot, skb);
1590 rxm->full_len -= pad;
1591 rxm->offset += prot->prepend_size;
1592 rxm->full_len -= prot->overhead_size;
1593 tls_advance_record_sn(sk, prot, &tls_ctx->rx);
1595 ctx->saved_data_ready(sk);
1603 int decrypt_skb(struct sock *sk, struct sk_buff *skb,
1604 struct scatterlist *sgout)
1609 return decrypt_internal(sk, skb, NULL, sgout, &chunk, &zc, false);
1612 static bool tls_sw_advance_skb(struct sock *sk, struct sk_buff *skb,
1615 struct tls_context *tls_ctx = tls_get_ctx(sk);
1616 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1619 struct strp_msg *rxm = strp_msg(skb);
1621 if (len < rxm->full_len) {
1623 rxm->full_len -= len;
1629 /* Finished with message */
1630 ctx->recv_pkt = NULL;
1631 __strp_unpause(&ctx->strp);
1636 /* This function traverses the rx_list in tls receive context to copies the
1637 * decrypted records into the buffer provided by caller zero copy is not
1638 * true. Further, the records are removed from the rx_list if it is not a peek
1639 * case and the record has been consumed completely.
1641 static int process_rx_list(struct tls_sw_context_rx *ctx,
1650 struct sk_buff *skb = skb_peek(&ctx->rx_list);
1653 struct tls_msg *tlm;
1656 /* Set the record type in 'control' if caller didn't pass it */
1659 ctrl = tlm->control;
1662 while (skip && skb) {
1663 struct strp_msg *rxm = strp_msg(skb);
1666 /* Cannot process a record of different type */
1667 if (ctrl != tlm->control)
1670 if (skip < rxm->full_len)
1673 skip = skip - rxm->full_len;
1674 skb = skb_peek_next(skb, &ctx->rx_list);
1677 while (len && skb) {
1678 struct sk_buff *next_skb;
1679 struct strp_msg *rxm = strp_msg(skb);
1680 int chunk = min_t(unsigned int, rxm->full_len - skip, len);
1684 /* Cannot process a record of different type */
1685 if (ctrl != tlm->control)
1688 /* Set record type if not already done. For a non-data record,
1689 * do not proceed if record type could not be copied.
1692 int cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1693 sizeof(ctrl), &ctrl);
1695 if (ctrl != TLS_RECORD_TYPE_DATA) {
1696 if (cerr || msg->msg_flags & MSG_CTRUNC)
1703 if (!zc || (rxm->full_len - skip) > len) {
1704 int err = skb_copy_datagram_msg(skb, rxm->offset + skip,
1711 copied = copied + chunk;
1713 /* Consume the data from record if it is non-peek case*/
1715 rxm->offset = rxm->offset + chunk;
1716 rxm->full_len = rxm->full_len - chunk;
1718 /* Return if there is unconsumed data in the record */
1719 if (rxm->full_len - skip)
1723 /* The remaining skip-bytes must lie in 1st record in rx_list.
1724 * So from the 2nd record, 'skip' should be 0.
1729 msg->msg_flags |= MSG_EOR;
1731 next_skb = skb_peek_next(skb, &ctx->rx_list);
1734 skb_unlink(skb, &ctx->rx_list);
1745 int tls_sw_recvmsg(struct sock *sk,
1752 struct tls_context *tls_ctx = tls_get_ctx(sk);
1753 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1754 struct tls_prot_info *prot = &tls_ctx->prot_info;
1755 struct sk_psock *psock;
1756 unsigned char control = 0;
1757 ssize_t decrypted = 0;
1758 struct strp_msg *rxm;
1759 struct tls_msg *tlm;
1760 struct sk_buff *skb;
1763 int target, err = 0;
1765 bool is_kvec = iov_iter_is_kvec(&msg->msg_iter);
1766 bool is_peek = flags & MSG_PEEK;
1767 bool bpf_strp_enabled;
1773 if (unlikely(flags & MSG_ERRQUEUE))
1774 return sock_recv_errqueue(sk, msg, len, SOL_IP, IP_RECVERR);
1776 psock = sk_psock_get(sk);
1778 bpf_strp_enabled = sk_psock_strp_enabled(psock);
1780 /* Process pending decrypted records. It must be non-zero-copy */
1781 err = process_rx_list(ctx, msg, &control, &cmsg, 0, len, false,
1784 tls_err_abort(sk, err);
1793 target = sock_rcvlowat(sk, flags & MSG_WAITALL, len);
1795 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
1797 while (len && (decrypted + copied < target || ctx->recv_pkt)) {
1798 bool retain_skb = false;
1805 skb = tls_wait_data(sk, psock, flags & MSG_DONTWAIT, timeo, &err);
1808 int ret = sk_msg_recvmsg(sk, psock, msg, len,
1820 if (prot->version == TLS_1_3_VERSION)
1823 tlm->control = ctx->control;
1826 rxm = strp_msg(skb);
1828 to_decrypt = rxm->full_len - prot->overhead_size;
1830 if (to_decrypt <= len && !is_kvec && !is_peek &&
1831 ctx->control == TLS_RECORD_TYPE_DATA &&
1832 prot->version != TLS_1_3_VERSION &&
1836 /* Do not use async mode if record is non-data */
1837 if (ctx->control == TLS_RECORD_TYPE_DATA && !bpf_strp_enabled)
1838 async_capable = ctx->async_capable;
1840 async_capable = false;
1842 err = decrypt_skb_update(sk, skb, &msg->msg_iter,
1843 &chunk, &zc, async_capable);
1844 if (err < 0 && err != -EINPROGRESS) {
1845 tls_err_abort(sk, -EBADMSG);
1849 if (err == -EINPROGRESS) {
1852 } else if (prot->version == TLS_1_3_VERSION) {
1853 tlm->control = ctx->control;
1856 /* If the type of records being processed is not known yet,
1857 * set it to record type just dequeued. If it is already known,
1858 * but does not match the record type just dequeued, go to end.
1859 * We always get record type here since for tls1.2, record type
1860 * is known just after record is dequeued from stream parser.
1861 * For tls1.3, we disable async.
1865 control = tlm->control;
1866 else if (control != tlm->control)
1872 cerr = put_cmsg(msg, SOL_TLS, TLS_GET_RECORD_TYPE,
1873 sizeof(control), &control);
1875 if (control != TLS_RECORD_TYPE_DATA) {
1876 if (cerr || msg->msg_flags & MSG_CTRUNC) {
1884 goto pick_next_record;
1887 if (bpf_strp_enabled) {
1888 err = sk_psock_tls_strp_read(psock, skb);
1889 if (err != __SK_PASS) {
1890 rxm->offset = rxm->offset + rxm->full_len;
1892 if (err == __SK_DROP)
1894 ctx->recv_pkt = NULL;
1895 __strp_unpause(&ctx->strp);
1900 if (rxm->full_len > len) {
1904 chunk = rxm->full_len;
1907 err = skb_copy_datagram_msg(skb, rxm->offset,
1913 rxm->offset = rxm->offset + chunk;
1914 rxm->full_len = rxm->full_len - chunk;
1925 /* For async or peek case, queue the current skb */
1926 if (async || is_peek || retain_skb) {
1927 skb_queue_tail(&ctx->rx_list, skb);
1931 if (tls_sw_advance_skb(sk, skb, chunk)) {
1932 /* Return full control message to
1933 * userspace before trying to parse
1934 * another message type
1936 msg->msg_flags |= MSG_EOR;
1937 if (control != TLS_RECORD_TYPE_DATA)
1946 /* Wait for all previously submitted records to be decrypted */
1947 spin_lock_bh(&ctx->decrypt_compl_lock);
1948 ctx->async_notify = true;
1949 pending = atomic_read(&ctx->decrypt_pending);
1950 spin_unlock_bh(&ctx->decrypt_compl_lock);
1952 err = crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
1954 /* one of async decrypt failed */
1955 tls_err_abort(sk, err);
1961 reinit_completion(&ctx->async_wait.completion);
1964 /* There can be no concurrent accesses, since we have no
1965 * pending decrypt operations
1967 WRITE_ONCE(ctx->async_notify, false);
1969 /* Drain records from the rx_list & copy if required */
1970 if (is_peek || is_kvec)
1971 err = process_rx_list(ctx, msg, &control, &cmsg, copied,
1972 decrypted, false, is_peek);
1974 err = process_rx_list(ctx, msg, &control, &cmsg, 0,
1975 decrypted, true, is_peek);
1977 tls_err_abort(sk, err);
1983 copied += decrypted;
1988 sk_psock_put(sk, psock);
1989 return copied ? : err;
1992 ssize_t tls_sw_splice_read(struct socket *sock, loff_t *ppos,
1993 struct pipe_inode_info *pipe,
1994 size_t len, unsigned int flags)
1996 struct tls_context *tls_ctx = tls_get_ctx(sock->sk);
1997 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
1998 struct strp_msg *rxm = NULL;
1999 struct sock *sk = sock->sk;
2000 struct sk_buff *skb;
2010 timeo = sock_rcvtimeo(sk, flags & SPLICE_F_NONBLOCK);
2012 from_queue = !skb_queue_empty(&ctx->rx_list);
2014 skb = __skb_dequeue(&ctx->rx_list);
2016 skb = tls_wait_data(sk, NULL, flags & SPLICE_F_NONBLOCK, timeo,
2019 goto splice_read_end;
2021 err = decrypt_skb_update(sk, skb, NULL, &chunk, &zc, false);
2023 tls_err_abort(sk, -EBADMSG);
2024 goto splice_read_end;
2028 /* splice does not support reading control messages */
2029 if (ctx->control != TLS_RECORD_TYPE_DATA) {
2031 goto splice_read_end;
2034 rxm = strp_msg(skb);
2036 chunk = min_t(unsigned int, rxm->full_len, len);
2037 copied = skb_splice_bits(skb, sk, rxm->offset, pipe, chunk, flags);
2039 goto splice_read_end;
2042 ctx->recv_pkt = NULL;
2043 __strp_unpause(&ctx->strp);
2045 if (chunk < rxm->full_len) {
2046 __skb_queue_head(&ctx->rx_list, skb);
2048 rxm->full_len -= len;
2055 return copied ? : err;
2058 bool tls_sw_sock_is_readable(struct sock *sk)
2060 struct tls_context *tls_ctx = tls_get_ctx(sk);
2061 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2062 bool ingress_empty = true;
2063 struct sk_psock *psock;
2066 psock = sk_psock(sk);
2068 ingress_empty = list_empty(&psock->ingress_msg);
2071 return !ingress_empty || ctx->recv_pkt ||
2072 !skb_queue_empty(&ctx->rx_list);
2075 static int tls_read_size(struct strparser *strp, struct sk_buff *skb)
2077 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2078 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2079 struct tls_prot_info *prot = &tls_ctx->prot_info;
2080 char header[TLS_HEADER_SIZE + MAX_IV_SIZE];
2081 struct strp_msg *rxm = strp_msg(skb);
2082 size_t cipher_overhead;
2083 size_t data_len = 0;
2086 /* Verify that we have a full TLS header, or wait for more data */
2087 if (rxm->offset + prot->prepend_size > skb->len)
2090 /* Sanity-check size of on-stack buffer. */
2091 if (WARN_ON(prot->prepend_size > sizeof(header))) {
2096 /* Linearize header to local buffer */
2097 ret = skb_copy_bits(skb, rxm->offset, header, prot->prepend_size);
2102 ctx->control = header[0];
2104 data_len = ((header[4] & 0xFF) | (header[3] << 8));
2106 cipher_overhead = prot->tag_size;
2107 if (prot->version != TLS_1_3_VERSION &&
2108 prot->cipher_type != TLS_CIPHER_CHACHA20_POLY1305)
2109 cipher_overhead += prot->iv_size;
2111 if (data_len > TLS_MAX_PAYLOAD_SIZE + cipher_overhead +
2116 if (data_len < cipher_overhead) {
2121 /* Note that both TLS1.3 and TLS1.2 use TLS_1_2 version here */
2122 if (header[1] != TLS_1_2_VERSION_MINOR ||
2123 header[2] != TLS_1_2_VERSION_MAJOR) {
2128 tls_device_rx_resync_new_rec(strp->sk, data_len + TLS_HEADER_SIZE,
2129 TCP_SKB_CB(skb)->seq + rxm->offset);
2130 return data_len + TLS_HEADER_SIZE;
2133 tls_err_abort(strp->sk, ret);
2138 static void tls_queue(struct strparser *strp, struct sk_buff *skb)
2140 struct tls_context *tls_ctx = tls_get_ctx(strp->sk);
2141 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2145 ctx->recv_pkt = skb;
2148 ctx->saved_data_ready(strp->sk);
2151 static void tls_data_ready(struct sock *sk)
2153 struct tls_context *tls_ctx = tls_get_ctx(sk);
2154 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2155 struct sk_psock *psock;
2157 strp_data_ready(&ctx->strp);
2159 psock = sk_psock_get(sk);
2161 if (!list_empty(&psock->ingress_msg))
2162 ctx->saved_data_ready(sk);
2163 sk_psock_put(sk, psock);
2167 void tls_sw_cancel_work_tx(struct tls_context *tls_ctx)
2169 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2171 set_bit(BIT_TX_CLOSING, &ctx->tx_bitmask);
2172 set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask);
2173 cancel_delayed_work_sync(&ctx->tx_work.work);
2176 void tls_sw_release_resources_tx(struct sock *sk)
2178 struct tls_context *tls_ctx = tls_get_ctx(sk);
2179 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2180 struct tls_rec *rec, *tmp;
2183 /* Wait for any pending async encryptions to complete */
2184 spin_lock_bh(&ctx->encrypt_compl_lock);
2185 ctx->async_notify = true;
2186 pending = atomic_read(&ctx->encrypt_pending);
2187 spin_unlock_bh(&ctx->encrypt_compl_lock);
2190 crypto_wait_req(-EINPROGRESS, &ctx->async_wait);
2192 tls_tx_records(sk, -1);
2194 /* Free up un-sent records in tx_list. First, free
2195 * the partially sent record if any at head of tx_list.
2197 if (tls_ctx->partially_sent_record) {
2198 tls_free_partial_record(sk, tls_ctx);
2199 rec = list_first_entry(&ctx->tx_list,
2200 struct tls_rec, list);
2201 list_del(&rec->list);
2202 sk_msg_free(sk, &rec->msg_plaintext);
2206 list_for_each_entry_safe(rec, tmp, &ctx->tx_list, list) {
2207 list_del(&rec->list);
2208 sk_msg_free(sk, &rec->msg_encrypted);
2209 sk_msg_free(sk, &rec->msg_plaintext);
2213 crypto_free_aead(ctx->aead_send);
2214 tls_free_open_rec(sk);
2217 void tls_sw_free_ctx_tx(struct tls_context *tls_ctx)
2219 struct tls_sw_context_tx *ctx = tls_sw_ctx_tx(tls_ctx);
2224 void tls_sw_release_resources_rx(struct sock *sk)
2226 struct tls_context *tls_ctx = tls_get_ctx(sk);
2227 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2229 kfree(tls_ctx->rx.rec_seq);
2230 kfree(tls_ctx->rx.iv);
2232 if (ctx->aead_recv) {
2233 kfree_skb(ctx->recv_pkt);
2234 ctx->recv_pkt = NULL;
2235 skb_queue_purge(&ctx->rx_list);
2236 crypto_free_aead(ctx->aead_recv);
2237 strp_stop(&ctx->strp);
2238 /* If tls_sw_strparser_arm() was not called (cleanup paths)
2239 * we still want to strp_stop(), but sk->sk_data_ready was
2242 if (ctx->saved_data_ready) {
2243 write_lock_bh(&sk->sk_callback_lock);
2244 sk->sk_data_ready = ctx->saved_data_ready;
2245 write_unlock_bh(&sk->sk_callback_lock);
2250 void tls_sw_strparser_done(struct tls_context *tls_ctx)
2252 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2254 strp_done(&ctx->strp);
2257 void tls_sw_free_ctx_rx(struct tls_context *tls_ctx)
2259 struct tls_sw_context_rx *ctx = tls_sw_ctx_rx(tls_ctx);
2264 void tls_sw_free_resources_rx(struct sock *sk)
2266 struct tls_context *tls_ctx = tls_get_ctx(sk);
2268 tls_sw_release_resources_rx(sk);
2269 tls_sw_free_ctx_rx(tls_ctx);
2272 /* The work handler to transmitt the encrypted records in tx_list */
2273 static void tx_work_handler(struct work_struct *work)
2275 struct delayed_work *delayed_work = to_delayed_work(work);
2276 struct tx_work *tx_work = container_of(delayed_work,
2277 struct tx_work, work);
2278 struct sock *sk = tx_work->sk;
2279 struct tls_context *tls_ctx = tls_get_ctx(sk);
2280 struct tls_sw_context_tx *ctx;
2282 if (unlikely(!tls_ctx))
2285 ctx = tls_sw_ctx_tx(tls_ctx);
2286 if (test_bit(BIT_TX_CLOSING, &ctx->tx_bitmask))
2289 if (!test_and_clear_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask))
2292 if (mutex_trylock(&tls_ctx->tx_lock)) {
2294 tls_tx_records(sk, -1);
2296 mutex_unlock(&tls_ctx->tx_lock);
2297 } else if (!test_and_set_bit(BIT_TX_SCHEDULED, &ctx->tx_bitmask)) {
2298 /* Someone is holding the tx_lock, they will likely run Tx
2299 * and cancel the work on their way out of the lock section.
2300 * Schedule a long delay just in case.
2302 schedule_delayed_work(&ctx->tx_work.work, msecs_to_jiffies(10));
2306 void tls_sw_write_space(struct sock *sk, struct tls_context *ctx)
2308 struct tls_sw_context_tx *tx_ctx = tls_sw_ctx_tx(ctx);
2310 /* Schedule the transmission if tx list is ready */
2311 if (is_tx_ready(tx_ctx) &&
2312 !test_and_set_bit(BIT_TX_SCHEDULED, &tx_ctx->tx_bitmask))
2313 schedule_delayed_work(&tx_ctx->tx_work.work, 0);
2316 void tls_sw_strparser_arm(struct sock *sk, struct tls_context *tls_ctx)
2318 struct tls_sw_context_rx *rx_ctx = tls_sw_ctx_rx(tls_ctx);
2320 write_lock_bh(&sk->sk_callback_lock);
2321 rx_ctx->saved_data_ready = sk->sk_data_ready;
2322 sk->sk_data_ready = tls_data_ready;
2323 write_unlock_bh(&sk->sk_callback_lock);
2325 strp_check_rcv(&rx_ctx->strp);
2328 int tls_set_sw_offload(struct sock *sk, struct tls_context *ctx, int tx)
2330 struct tls_context *tls_ctx = tls_get_ctx(sk);
2331 struct tls_prot_info *prot = &tls_ctx->prot_info;
2332 struct tls_crypto_info *crypto_info;
2333 struct tls12_crypto_info_aes_gcm_128 *gcm_128_info;
2334 struct tls12_crypto_info_aes_gcm_256 *gcm_256_info;
2335 struct tls12_crypto_info_aes_ccm_128 *ccm_128_info;
2336 struct tls12_crypto_info_chacha20_poly1305 *chacha20_poly1305_info;
2337 struct tls_sw_context_tx *sw_ctx_tx = NULL;
2338 struct tls_sw_context_rx *sw_ctx_rx = NULL;
2339 struct cipher_context *cctx;
2340 struct crypto_aead **aead;
2341 struct strp_callbacks cb;
2342 u16 nonce_size, tag_size, iv_size, rec_seq_size, salt_size;
2343 struct crypto_tfm *tfm;
2344 char *iv, *rec_seq, *key, *salt, *cipher_name;
2354 if (!ctx->priv_ctx_tx) {
2355 sw_ctx_tx = kzalloc(sizeof(*sw_ctx_tx), GFP_KERNEL);
2360 ctx->priv_ctx_tx = sw_ctx_tx;
2363 (struct tls_sw_context_tx *)ctx->priv_ctx_tx;
2366 if (!ctx->priv_ctx_rx) {
2367 sw_ctx_rx = kzalloc(sizeof(*sw_ctx_rx), GFP_KERNEL);
2372 ctx->priv_ctx_rx = sw_ctx_rx;
2375 (struct tls_sw_context_rx *)ctx->priv_ctx_rx;
2380 crypto_init_wait(&sw_ctx_tx->async_wait);
2381 spin_lock_init(&sw_ctx_tx->encrypt_compl_lock);
2382 crypto_info = &ctx->crypto_send.info;
2384 aead = &sw_ctx_tx->aead_send;
2385 INIT_LIST_HEAD(&sw_ctx_tx->tx_list);
2386 INIT_DELAYED_WORK(&sw_ctx_tx->tx_work.work, tx_work_handler);
2387 sw_ctx_tx->tx_work.sk = sk;
2389 crypto_init_wait(&sw_ctx_rx->async_wait);
2390 spin_lock_init(&sw_ctx_rx->decrypt_compl_lock);
2391 crypto_info = &ctx->crypto_recv.info;
2393 skb_queue_head_init(&sw_ctx_rx->rx_list);
2394 aead = &sw_ctx_rx->aead_recv;
2397 switch (crypto_info->cipher_type) {
2398 case TLS_CIPHER_AES_GCM_128: {
2399 nonce_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2400 tag_size = TLS_CIPHER_AES_GCM_128_TAG_SIZE;
2401 iv_size = TLS_CIPHER_AES_GCM_128_IV_SIZE;
2402 iv = ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->iv;
2403 rec_seq_size = TLS_CIPHER_AES_GCM_128_REC_SEQ_SIZE;
2405 ((struct tls12_crypto_info_aes_gcm_128 *)crypto_info)->rec_seq;
2407 (struct tls12_crypto_info_aes_gcm_128 *)crypto_info;
2408 keysize = TLS_CIPHER_AES_GCM_128_KEY_SIZE;
2409 key = gcm_128_info->key;
2410 salt = gcm_128_info->salt;
2411 salt_size = TLS_CIPHER_AES_GCM_128_SALT_SIZE;
2412 cipher_name = "gcm(aes)";
2415 case TLS_CIPHER_AES_GCM_256: {
2416 nonce_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2417 tag_size = TLS_CIPHER_AES_GCM_256_TAG_SIZE;
2418 iv_size = TLS_CIPHER_AES_GCM_256_IV_SIZE;
2419 iv = ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->iv;
2420 rec_seq_size = TLS_CIPHER_AES_GCM_256_REC_SEQ_SIZE;
2422 ((struct tls12_crypto_info_aes_gcm_256 *)crypto_info)->rec_seq;
2424 (struct tls12_crypto_info_aes_gcm_256 *)crypto_info;
2425 keysize = TLS_CIPHER_AES_GCM_256_KEY_SIZE;
2426 key = gcm_256_info->key;
2427 salt = gcm_256_info->salt;
2428 salt_size = TLS_CIPHER_AES_GCM_256_SALT_SIZE;
2429 cipher_name = "gcm(aes)";
2432 case TLS_CIPHER_AES_CCM_128: {
2433 nonce_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2434 tag_size = TLS_CIPHER_AES_CCM_128_TAG_SIZE;
2435 iv_size = TLS_CIPHER_AES_CCM_128_IV_SIZE;
2436 iv = ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->iv;
2437 rec_seq_size = TLS_CIPHER_AES_CCM_128_REC_SEQ_SIZE;
2439 ((struct tls12_crypto_info_aes_ccm_128 *)crypto_info)->rec_seq;
2441 (struct tls12_crypto_info_aes_ccm_128 *)crypto_info;
2442 keysize = TLS_CIPHER_AES_CCM_128_KEY_SIZE;
2443 key = ccm_128_info->key;
2444 salt = ccm_128_info->salt;
2445 salt_size = TLS_CIPHER_AES_CCM_128_SALT_SIZE;
2446 cipher_name = "ccm(aes)";
2449 case TLS_CIPHER_CHACHA20_POLY1305: {
2450 chacha20_poly1305_info = (void *)crypto_info;
2452 tag_size = TLS_CIPHER_CHACHA20_POLY1305_TAG_SIZE;
2453 iv_size = TLS_CIPHER_CHACHA20_POLY1305_IV_SIZE;
2454 iv = chacha20_poly1305_info->iv;
2455 rec_seq_size = TLS_CIPHER_CHACHA20_POLY1305_REC_SEQ_SIZE;
2456 rec_seq = chacha20_poly1305_info->rec_seq;
2457 keysize = TLS_CIPHER_CHACHA20_POLY1305_KEY_SIZE;
2458 key = chacha20_poly1305_info->key;
2459 salt = chacha20_poly1305_info->salt;
2460 salt_size = TLS_CIPHER_CHACHA20_POLY1305_SALT_SIZE;
2461 cipher_name = "rfc7539(chacha20,poly1305)";
2469 /* Sanity-check the sizes for stack allocations. */
2470 if (iv_size > MAX_IV_SIZE || nonce_size > MAX_IV_SIZE ||
2471 rec_seq_size > TLS_MAX_REC_SEQ_SIZE) {
2476 if (crypto_info->version == TLS_1_3_VERSION) {
2478 prot->aad_size = TLS_HEADER_SIZE;
2479 prot->tail_size = 1;
2481 prot->aad_size = TLS_AAD_SPACE_SIZE;
2482 prot->tail_size = 0;
2485 prot->version = crypto_info->version;
2486 prot->cipher_type = crypto_info->cipher_type;
2487 prot->prepend_size = TLS_HEADER_SIZE + nonce_size;
2488 prot->tag_size = tag_size;
2489 prot->overhead_size = prot->prepend_size +
2490 prot->tag_size + prot->tail_size;
2491 prot->iv_size = iv_size;
2492 prot->salt_size = salt_size;
2493 cctx->iv = kmalloc(iv_size + salt_size, GFP_KERNEL);
2498 /* Note: 128 & 256 bit salt are the same size */
2499 prot->rec_seq_size = rec_seq_size;
2500 memcpy(cctx->iv, salt, salt_size);
2501 memcpy(cctx->iv + salt_size, iv, iv_size);
2502 cctx->rec_seq = kmemdup(rec_seq, rec_seq_size, GFP_KERNEL);
2503 if (!cctx->rec_seq) {
2509 *aead = crypto_alloc_aead(cipher_name, 0, 0);
2510 if (IS_ERR(*aead)) {
2511 rc = PTR_ERR(*aead);
2517 ctx->push_pending_record = tls_sw_push_pending_record;
2519 rc = crypto_aead_setkey(*aead, key, keysize);
2524 rc = crypto_aead_setauthsize(*aead, prot->tag_size);
2529 tfm = crypto_aead_tfm(sw_ctx_rx->aead_recv);
2531 if (crypto_info->version == TLS_1_3_VERSION)
2532 sw_ctx_rx->async_capable = 0;
2534 sw_ctx_rx->async_capable =
2535 !!(tfm->__crt_alg->cra_flags &
2538 /* Set up strparser */
2539 memset(&cb, 0, sizeof(cb));
2540 cb.rcv_msg = tls_queue;
2541 cb.parse_msg = tls_read_size;
2543 strp_init(&sw_ctx_rx->strp, sk, &cb);
2549 crypto_free_aead(*aead);
2552 kfree(cctx->rec_seq);
2553 cctx->rec_seq = NULL;
2559 kfree(ctx->priv_ctx_tx);
2560 ctx->priv_ctx_tx = NULL;
2562 kfree(ctx->priv_ctx_rx);
2563 ctx->priv_ctx_rx = NULL;